Conditional Imaging: camera optical state per frame conceals the scene until authorized reconstructionarxiv.orgOctober 22, 2026LabOn deckBriefed Oct 22
PRAISE: query-conditioned attribute subspaces lift VLM ReID at test time, no ID labelsarxiv.orgOctober 22, 2026LabOn deckBriefed Oct 22
Register Anything Model: one 6-DoF point-cloud registrar across indoor, outdoor, and object LiDARarxiv.orgOctober 22, 2026LabOn deckBriefed Oct 22
RGB–IR fusion for counter-UAV: early fusion wins raw mAP; late fusion survives misaligned sensorsarxiv.orgOctober 22, 2026LabOn deckBriefed Oct 22
CodecMotion-OAD: decoder motion vectors replace optical flow for streaming action detection; 11.48× CPUarxiv.orgOctober 21, 2026LabOn deckBriefed Oct 21
Foreground Attraction: weakly supervised camera-only 3D occupancy; foreground IoU 6.29 → 6.96arxiv.orgOctober 21, 2026LabOn deckBriefed Oct 21
Jetson AGX Orin depth bake-off: NVIDIA ESS 0.70% AbsRel @ 8.6 ms; monocular models scatter 17–46% of edgesarxiv.orgOctober 21, 2026LabOn deckBriefed Oct 21
CGPLR: unsupervised cross-domain ReID; 76.4% Rank-1 Market→MSMT17, 95.1% MSMT17→Marketarxiv.orgOctober 20, 2026LabOn deckBriefed Oct 20
EMMA: event-driven multimodal memory for hours-long video; dual-stream boundaries, not fixed clipsarxiv.orgOctober 20, 2026LabOn deckBriefed Oct 20
GeoForward: sparse LiDAR anchors feedforward 3D Gaussian splatting; lowest nuScenes cross-camera depth errorarxiv.orgOctober 20, 2026LabOn deckBriefed Oct 20
YUTELA: Jetson Thor multi-site VAD; 31 CCTV cams / 170 h; TensorRT 14.17 → 29.94 FPSarxiv.orgOctober 20, 2026LabOn deckBriefed Oct 20
ODSS-Net: orthogonal shared-structure / modality-specific ReID so alignment does not erase the extra cameraarxiv.orgOctober 19, 2026LabOn deckBriefed Oct 19
RCA-MOT: RGB–infrared MOT that does not crush both cameras into one vector; +5.4 / +6.1 HOTAarxiv.orgOctober 19, 2026LabOn deckBriefed Oct 19
TriSAR-3D / PaintFusion: LiDAR + RGB + thermal 3D person detect; 65.4 AP@50, 51.6 ms; lying-pose models collapse without lying trainingarxiv.orgOctober 19, 2026LabOn deckBriefed Oct 19
ARC-SORT: uncertainty-adaptive MOT; 81.82 HOTA SportsMOT, fewest ID switches of 69 entriesarxiv.orgOctober 16, 2026LabOn deckBriefed Oct 16
FALCON: KR260 FPGA occupancy with a bounded tail; Jetson Orin 4.4× median, L4 23.6×arxiv.orgOctober 16, 2026LabOn deckBriefed Oct 16
MaLoW: streaming VLM that writes history into LoRA weights, not the context windowarxiv.orgOctober 16, 2026LabOn deckBriefed Oct 16
DeltaStreamer: agentic streaming video with change-aware memory + targeted replayarxiv.orgOctober 15, 2026LabOn deckBriefed Oct 15
Detector architecture vs scene density: RT-DETR wins crowded MOT; YOLO26-L + NMS is the speed/quality splitarxiv.orgOctober 15, 2026LabOn deckBriefed Oct 15
LidarVerse: NVIDIA SIL real-time unified LiDAR world model; 11 FPS on H100arxiv.orgOctober 15, 2026LabOn deckBriefed Oct 15
SAM3RL: RL memory controller for SAM 3 tracking; SOTA on 10 of 12 benches, no runtime taxarxiv.orgOctober 15, 2026LabOn deckBriefed Oct 15
InfraWorld: roadside 4D occupancy as selective state evolutionarxiv.orgOctober 14, 2026LabOn deckBriefed Oct 14
SPARE: streaming VLM that gates history retrieval and shares the visual encoderarxiv.orgOctober 14, 2026LabOn deckBriefed Oct 14
Temporal Evidence Selection: VAD explanations need object-tag–conditioned sampling, not anomaly saliencyarxiv.orgOctober 14, 2026LabOn deckBriefed Oct 14
When Trackers Fail: VLM Dual-Agent verifies SAM-class trackers and recovers identity driftarxiv.orgOctober 14, 2026LabOn deckBriefed Oct 14
Event cameras for long action sequences: multi-frame MAE beats single-frame event stacksarxiv.orgOctober 13, 2026LabOn deckBriefed Oct 13
MemReg: streaming outdoor LiDAR registration with hybrid pose memory (NeurIPS 2026)arxiv.orgOctober 13, 2026LabOn deckBriefed Oct 13
SpatialASM: train LVLMs to revisit the scene from task-relevant viewpointsarxiv.orgOctober 13, 2026LabOn deckBriefed Oct 13
AMGT: UAV maritime visible+IR+AIS vessel ID/track; MAir-YOLOv8-OBB + MAir-SORTarxiv.orgOctober 12, 2026LabOn deckBriefed Oct 12
OccludeBench: VLMs miss what’s hidden; 42.7% on cue-based occlusion vs 62.7% inferentialarxiv.orgOctober 12, 2026LabOn deckBriefed Oct 12
UniMergeTrack: merge RGB-T / RGB-Event / RGB-Depth trackers instead of joint-training themarxiv.orgOctober 12, 2026LabOn deckBriefed Oct 12
FastBench: streaming VLMs fail high-dynamic streams; Gemini-3.5-Flash 50.7%, 2 FPS misses the eventarxiv.orgOctober 9, 2026LabOn deckBriefed Oct 9
Onboard maritime YOLOX-S on Versal VC1902: tiles keep 98% of vessels at 29% of scene volumearxiv.orgOctober 9, 2026LabOn deckBriefed Oct 9
BagDINO: DINOv3 + LoRA multi-view baggage ReID for mishandled airport bagsarxiv.orgOctober 8, 2026LabOn deckBriefed Oct 8
arXiv: CCDF — deepfake surveillance benchmark; SOTA detectors fail on Grok / VEO / Sora crime clipsarxiv.orgOctober 7, 2026LabOn deckBriefed Oct 7
arXiv: frozen VideoLLMs already encode evidence readiness for streaming QAarxiv.orgOctober 7, 2026LabOn deckBriefed Oct 7
arXiv: RBA — explainable weakly-supervised video anomaly detection on MLLMsarxiv.orgOctober 6, 2026LabOn deckBriefed Oct 6
arXiv: TRIM-ReID — RGB + NIR + thermal ReID with duplication-aware token droparxiv.orgOctober 6, 2026LabOn deckBriefed Oct 6
arXiv: referring MOT on a moving camera — residual motion after ego-motion subtractarxiv.orgOctober 6, 2026LabOn deckBriefed Oct 6
arXiv: CHASE-VLA — post-training quantization for vision-language-action modelsarxiv.orgOctober 5, 2026LabOn deckBriefed Oct 5
arXiv: FORESIGHT — training-free “plan the next look” for streaming VLMsarxiv.orgOctober 5, 2026LabOn deckBriefed Oct 5
arXiv: Physical AI Smart Spaces — NVIDIA multi-camera 3D bench, ~1,800 cameras, 280+ hoursarxiv.orgOctober 5, 2026LabOn deckBriefed Oct 5
arXiv: ASH — zero-shot open-vocab MOT/segmentation on long video, <25 GB GPUarxiv.orgOctober 2, 2026LabOn deckBriefed Oct 2
arXiv: FlashBack — training-free “when to remember” for streaming VLMsarxiv.orgOctober 2, 2026LabOn deckBriefed Oct 2
arXiv: Video Evidence Indexing — cheap preview, then spend the high-res token budget on the moments that matterarxiv.orgOctober 2, 2026LabOn deckBriefed Oct 2
arXiv: GaugeVLM — measured 3D interventions fix spatial VLM self-contradictionarxiv.orgOctober 1, 2026LabOn deckBriefed Oct 1
arXiv: PhyProbe — physical-consistency scorer for generated video (NeurIPS 2026 poster)arxiv.orgOctober 1, 2026LabOn deckBriefed Oct 1
arXiv: Video-RSI — video agents rewrite their own evidence harnessarxiv.orgSeptember 30, 2026LabOn deckBriefed Sep 30
arXiv: VideoLoop — rewrite working memory so long-video agents stop thrashingarxiv.orgSeptember 30, 2026LabOn deckBriefed Sep 30
arXiv: EdgeVLN — 4-bit VLN on Jetson Orin NX 16 GB, 20.8× faster than streamed BF16arxiv.orgSeptember 29, 2026LabOn deckBriefed Sep 29
arXiv: Rift — Jetson AGX Orin VLM, −78% energy / −69% latency by dropping answer-invariant tilesarxiv.orgSeptember 28, 2026LabOn deckBriefed Sep 28
arXiv: TrafficFab — edge–cloud fabric for ~400 live CCTV streams (10% of Bangalore) on Jetson + fogarxiv.orgSeptember 28, 2026LabOn deckBriefed Sep 28
arXiv: CDKF-Track — cooperative 3D MOT for EdgeAI, up to +28% vs multi-agent SOTAarxiv.orgSeptember 25, 2026LabOn deckBriefed Sep 25
arXiv: GTR — softmax-free dense backbone, TensorRT on DRIVE AGX Thor in 2.3–8.8 msarxiv.orgSeptember 25, 2026LabOn deckBriefed Sep 25
arXiv: LEAP-NBV — foundation-model next-best-view on a Jetson Xavier NX at 3.6 fps / 2.6 Jarxiv.orgSeptember 23, 2026LabOn deckBriefed Sep 23
arXiv: FIVE-VLA — 641M driving VLA, ~4 fps on a T4 (edge proxy), 8–30× fasterarxiv.orgSeptember 22, 2026LabOn deckBriefed Sep 22
arXiv: RoboDawn — agentic VLM drives a robot with translate / rotate / grip; one demo → SOTAarxiv.orgSeptember 22, 2026LabOn deckBriefed Sep 22
arXiv: TERMon — FPGA runtime monitor on edge AI, two-cycle decision, no BRAM/DSParxiv.orgSeptember 22, 2026LabOn deckBriefed Sep 22
arXiv: PROVIA — online mistake detection on full egocentric trials, 58–70 fpsarxiv.orgSeptember 18, 2026LabOn deckBriefed Sep 18
arXiv: Real-Time EXPO-FT — VLA proposes, a fast editor reacts; 42% → 97% in 10 minutes of robot dataarxiv.orgSeptember 18, 2026LabOn deckBriefed Sep 18
arXiv: rMuscle — VLA “muscle memory” cache, 1.29–1.42× on Jetson Thorarxiv.orgSeptember 17, 2026LabOn deckBriefed Sep 17
arXiv: Tracking the Unseen — YOLOv11n + Kalman + ReID through full/long occlusionarxiv.orgSeptember 17, 2026LabOn deckBriefed Sep 17
arXiv: AVNet — one model classifies emergency vehicles from audio, video, or botharxiv.orgSeptember 16, 2026LabOn deckBriefed Sep 16
arXiv: PhysStream — streaming physics-grounded video with online scene memory + object tracksarxiv.orgSeptember 16, 2026LabOn deckBriefed Sep 16
arXiv: BVB / Blender-VideoBench — if the agent understood the video, it can rebuild it in Blenderarxiv.orgSeptember 15, 2026LabOn deckBriefed Sep 15
arXiv: PhysBrain 1.5 — 8B VLM → physical foundation model (understand / act / predict)arxiv.orgSeptember 15, 2026LabOn deckBriefed Sep 15
arXiv: CFD — caption once on the edge, pull frames only when the question needs pixelsarxiv.orgSeptember 14, 2026LabOn deckBriefed Sep 14
arXiv: EVPeriscope — ground robot’s event camera tracks a drone by its propellersarxiv.orgSeptember 14, 2026LabOn deckBriefed Sep 14
arXiv: FARM — 34k-param readout on a frozen robot world model flags failurearxiv.orgSeptember 14, 2026LabOn deckBriefed Sep 14
arXiv: MMGait — RGB + IR + depth + LiDAR + radar gait bench; one model, missing sensorsarxiv.orgSeptember 11, 2026LabOn deckBriefed Sep 11
arXiv: Point4D — feed-forward 4D tracks across 200+ frames, not a 30-frame windowarxiv.orgSeptember 11, 2026LabOn deckBriefed Sep 11
arXiv: LogiScope-VQA — warehouse VLM hazard bench; GPT-class models still miss humanarxiv.orgSeptember 10, 2026LabOn deckBriefed Sep 10
arXiv: TempoGround — streaming VLM grounding (enter / stay / leave) then 3D boxarxiv.orgSeptember 10, 2026LabOn deckBriefed Sep 10
arXiv: VLX-VR — agentic video reasoning with a Think–Memory–Observation looparxiv.orgSeptember 10, 2026LabOn deckBriefed Sep 10
AGC-VLN: training-free UAV + UGV collab via a shared bird’s-eye maparxiv.orgSeptember 9, 2026LabOn deckBriefed Sep 9
Antioch raises $32M Series A (Greylock) — sim platform on NVIDIA Isaac, customer Ringmorningstar.comSeptember 9, 2026LabOn deckBriefed Sep 9
arXiv: MINT — world-space camera + hand motion from egocentric RGB (physical AI)arxiv.orgSeptember 9, 2026LabOn deckBriefed Sep 9
arXiv: ReactVAU — streaming anomaly understanding, heavy VLM only when the fast path firesarxiv.orgSeptember 9, 2026LabOn deckBriefed Sep 9
arXiv: VLM as a tracker tuner — no target labels, recovers most of the MOT17→MOT20 droparxiv.orgSeptember 8, 2026LabOn deckBriefed Sep 8
arXiv: weakly supervised video anomaly detection at event level, not fixed top-k snippetsarxiv.orgSeptember 8, 2026LabOn deckBriefed Sep 8